pardnchiu/go-llm-router

Last updated 2026-10-06

[!NOTE] This README was generated by SKILL, get the ZH version from here.

ONE AGENT INTERFACE FOR EVERY LLM PROVIDER

A Go LLM library with a unified LLM API for 14 providers, provider@model routing, and normalized streaming and usage

Extracted as a standalone package from Agenvoy 2741d4a

Since v0.28.17, Agenvoy's LLM providers are unified and replaced by this package go-llm-router.

Table of Contents

• Features
• Architecture
• License
• Author

Features

go get github.com/pardnchiu/go-llm-router · Documentation

• Unified Agent Interface — Fourteen providers share one Send / SendStream contract, so callers never write per-vendor adapters.
• String Routing Factory — Resolve an Agent from a provider@model name; an unknown prefix falls through to an OpenAI-compatible endpoint as a named instance.
• Reasoning Level Normalization — Six levels from none to max map onto Claude thinking, Gemini budgets, and OpenAI effort, clamped to each model's range.
• Usage Field Absorption — prompt_tokens / input_tokens and every cache variant fold into a single Input / Output / Cache shape.
• Multimodal and OAuth Extensions — Built-in Copilot / Codex / Grok OAuth flows, plus image generation, speech-to-text, and text-to-speech agents.

Architecture

Full Architecture

graph TB
    App[Caller] --> Router[router.New]
    Router --> Agent[llmrouter.Agent]
    Agent --> KeyBased[Key-based providers]
    Agent --> OAuthBased[OAuth providers]
    Agent --> Compat[OpenAI-compatible endpoints]
    OAuthBased --> OAuth[core/oauth token store]
    Agent --> Stream[Stream event normalization]
    Agent --> Usage[Usage normalization]
    Agent --> Media[Image / audio agents]

License

This project is licensed under the MIT LICENSE.

Author

Just open an issue to share an idea.

©️ 2026 邱敬幃 Pardn Chiu